Labelbox Raises $100 Million Series E as the Reinforcement-Learning Layer Behind Frontier AI Models Turns Profitable

Labelbox, a company that sits in the unglamorous but indispensable layer between raw compute and capable models, has raised more than $100 million in Series E financing. The round, which includes equity and a credit facility, was led by Infinitum Partners with participation from Wipro Ventures, Andreessen Horowitz, B Capital, Tau Ventures and Snowpoint Ventures, the company announced Wednesday.

The pitch reflects how the AI economy has stratified. While labs race to deploy frontier models, a parallel market has grown up around making those models better after pretraining — and, increasingly, around making enterprise agents that actually improve with use. Labelbox operates on both sides of that divide with two core products: Horizon, which provides reinforcement-learning environments and evaluation infrastructure for frontier labs, and Recursion Managed Agents, which lets enterprises deploy fleets of agents inside existing business applications.

Horizon's environments simulate complex, multi-step work across coding, research and other demanding tasks, letting models practice, receive expert feedback, and improve through reinforcement learning. The company says it works with leading AI labs and hyperscalers on post-training programs — and, per company figures, serves more than 400 customers including over 90% of leading U.S. AI labs. Those are company-disclosed numbers, not independently audited.

The enterprise side is newer and, arguably, the growth engine. Recursion Managed Agents, opened to enterprise customers earlier this month, deploys agents that respond to schedules, alerts and tickets, coordinate across tools, and get their output evaluated against defined standards — with each run feeding the agent's memory so its future attempts improve. Labelbox itself uses more than 50 such agents internally across engineering, research, product, sales, marketing and finance.

The financial claims are notable for a company in this layer: Labelbox says it reached profitability in 2026 while growing more than fivefold year over year. If accurate, that puts it in rare company — most businesses riding the AI wave are spending venture capital far faster than they book revenue, and the ones claiming profits tend to be selling the picks and shovels.

"Every company will soon run on AI systems that learn from their own work," CEO and co-founder Manu Sharma said in the announcement. "Getting there takes innovation at every layer, from sandbox infrastructure and the agent harness to reinforcement learning post-training. Labelbox brings those layers together in one platform."

The funding will go toward research and engineering across Horizon and Recursion, plus team expansion. The strategic read: as frontier labs shift spending from raw pretraining compute toward post-training and evaluation, and as enterprises shift from pilots toward agent fleets judged by measurable outcomes, the infrastructure that measures and improves model behavior is becoming a market of its own — one that Labelbox just banked nine figures to defend.